Ok, after seeing this post by PJ Hyett [1], I have decided to skip to the end and go with Git [2].
So what I need is a beginner's practical guide to Git. "Beginner" being defined as someone who knows how to handle their compiler, understands to some level what a Makefile [3] is, and has touched source control without understanding it very well.
"Practical" being defined as this person doesn't want to get into great detail regarding what Git is doing in the background, and doesn't even care (or know) that it's distributed. Your answers might hint at the possibilities, but try to aim for the beginner that wants to keep a 'main' repository on a 'server' which is backed up and secure, and treat their local repository as merely a 'client' resource.
So:
I will go through the entries from time to time and 'tidy' them up so they have a consistent look/feel and it's easy to scan the list - feel free to follow a simple "header - brief explanation - list of instructions - gotchas and extra info" template. I'll also link to the entries from the bullet list above so it's easy to find them later.
A git repository is simply a directory containing a special .git
directory.
This is different from "centralised" version-control systems (like subversion), where a "repository" is hosted on a remote server, which you checkout
into a "working copy" directory. With git, your working copy is the repository.
Simply run git init
in the directory which contains the files you wish to track.
For example,
cd ~/code/project001/
git init
This creates a .git
(hidden) folder in the current directory.
To make a new project, run git init
with an additional argument (the name of the directory to be created):
git init project002
(This is equivalent to: mkdir project002 && cd project002 && git init)
To check if the current current path is within a git repository, simply run git status
- if it's not a repository, it will report "fatal: Not a git repository"
You could also list the .git
directory, and check it contains files/directories similar to the following:
$ ls .git
HEAD config hooks/ objects/
branches/ description info/ refs/
If for whatever reason you wish to "de-git" a repository (you wish to stop using git to track that project). Simply remove the .git
directory at the base level of the repository.
cd ~/code/project001/
rm -rf .git/
Caution: This will destroy all revision history, all your tags, everything git has done. It will not touch the "current" files (the files you can currently see), but previous changes, deleted files and so on will be unrecoverable!
Included with git — Run git gui
from the command line, and the Windows
msysgit
[1] installer adds it to the Start menu.
Git GUI can do a majority of what you'd need to do with git. Including stage changes, configure git and repositories, push changes, create/checkout/delete branches, merge, and many other things.
One of my favourite features is the "stage line" and "stage hunk" shortcuts in the right-click menu, which lets you commit specific parts of a file. You can achieve the same via git add -i
, but I find it easier to use.
It isn't the prettiest application, but it works on almost all platforms (being based upon Tcl/Tk)
Screenshots [2] | a screencast [3]
Also included with git. It is a git history viewer, and lets you visualise a repository's history (including branches, when they are created, and merged). You can view and search commits.
Goes together nicely with git-gui.
Mac OS X application. Mainly an equivalent of git log
, but has some integration with
github
[6] (like the "Network view").
Looks pretty, and fits with Mac OS X. You can search repositories. The biggest critisism of Gitnub is that it shows history in a linear fashion (a single branch at a time) - it doesn't visualise branching and merging, which can be important with git, although this is a planned improvement.
Download links, change log and screenshots [7] | git repository [8]
Intends to be a "gitk clone for OS X".
It can visualise non-linear branching history, perform commits, view and search commits, and it has some other nice features like being able to "Quicklook" any file in any revision (press space in the file-list view), export any file (via drag and drop).
It is far better integrated into OS X than git-gui
/gitk
, and is fast and stable even with exceptionally large repositories.
The original git repository pieter [10] has not updated recently (over a year at time of writing). A more actively maintained branch is available at brotherbard/gitx [11] - it adds "sidebar, fetch, pull, push, add remote, merge, cherry-pick, rebase, clone, clone to"
Download [12] | Screenshots [13] | git repository [14] | brotherbard fork [15] | laullon fork [16]
From the homepage:
SmartGit is a front-end for the distributed version control system Git and runs on Windows, Mac OS X and Linux. SmartGit is intended for developers who prefer a graphical user interface over a command line client, to be even more productive with Git — the most powerful DVCS today.
You can download it from their website [18].
Download [19]
TortoiseSVN Git version for Windows users.
It is porting TortoiseSVN to TortoiseGit The latest release 1.2.1.0 This release can complete regular task, such commit, show log, diff two version, create branch and tag, Create patch and so on. See ReleaseNotes [21] for detail. Welcome to contribute this project.
Download [22]
QGit is a git GUI viewer built on Qt/C++.
With qgit you will be able to browse revisions history, view patch content and changed files, graphically following different development branches.
Download [24]
gitg is a git repository viewer targeting gtk+/GNOME. One of its main objectives is to provide a more unified user experience for git frontends across multiple desktops. It does this not be writing a cross-platform application, but by close collaboration with similar clients for other operating systems (like GitX for OS X).
Download: releases [26] or source [27]
Gitbox is a Mac OS X graphical interface for Git version control system. In a single window you see branches, history and working directory status.
Everyday operations are easy: stage and unstage changes with a checkbox. Commit, pull, merge and push with a single click. Double-click a change to show a diff with FileMerge.app.
Download [29]
The Gity website doesn't have much information, but from the screenshots on there it appears to be a feature rich open source OS X git gui.
Meld is a visual diff and merge tool. You can compare two or three files and edit them in place (diffs update dynamically). You can compare two or three folders and launch file comparisons. You can browse and view a working copy from popular version control systems such such as CVS, Subversion, Bazaar-ng and Mercurial [and Git].
Downloads [34]
A Git GUIfor OSX by Steve Dekorte.
At a glance, see which remote branches have changes to pull and local repos have changes to push. The git ops of add, commit, push, pull, tag and reset are supported as well as visual diffs and visual browsing of project hieracy that highlights local changes and additions.
Free for 1 repository, $25 for more.
Download [36]
Focuses on making Git easy to use. Features a native Cocoa (mac-like) UI, fast repository browsing, cloning, push/pull, branching/merging, visual diff, remote branches, easy access to the Terminal, and more.
By making the most commonly used Git actions intuitive and easy to perform, Sprout (formerly GitMac) makes Git user-friendly. Compatible with most Git workflows, Sprout is great for designers and developers, team collaboration and advanced and novice users alike.
A feature-rich Git GUI for Mac OSX. 30-day free trial, $59USD for a single-user license.
EGit is an Eclipse Team provider for the Git version control system. Git is a distributed SCM, which means every developer has a full copy of all history of every revision of the code, making queries against the history very fast and versatile.
The EGit project is implementing Eclipse tooling on top of the JGit Java implementation of Git.
Open Source for Windows - installs everything you need to work with Git in a single package, easy to use.
Git Extensions is a toolkit to make working with Git on Windows more intuitive. The shell extension will intergrate in Windows Explorer and presents a context menu on files and directories. There is also a Visual Studio plugin to use git from Visual Studio.
Download [47]
Big thanks to dbr [48] for elaborating on the git gui stuff.
SourceTree is a free Mac client for Git, Mercurial and SVN. Built by Atlassian, the folks behind BitBucket, it seems to work equally well with any VC system, which allows you to master a single tool for use with all of your projects, however they're version-controlled. Feature-packed, and FREE.
Expert-Ready & Feature-packed for both novice and advanced users:
Review outgoing and incoming changesets. Cherry-pick between branches. Patch handling, rebase, stash / shelve and much more.
Well, despite the fact that you asked that we not "simply" link to other resources, it's pretty foolish when there already exists a community grown (and growing) resource that's really quite good: the Git Community Book [1]. Seriously, this 20+ questions in a question is going to be anything but concise and consistent. The Git Community Book is available as both HTML and PDF and answers many of your questions with clear, well formatted and peer reviewed answers and in a format that allows you to jump straight to your problem at hand.
Alas, if my post really upsets you then I'll delete it. Just say so.
[1] http://book.git-scm.com/The ability to have git ignore files you don't wish it to track is very useful.
To ignore a file or set of files you supply a pattern. The pattern syntax for git is fairly simple, but powerful. It is applicable to all three of the different files I will mention bellow.
Great Example from the gitignore(5) [1] man page:
$ git status
[...]
# Untracked files:
[...]
# Documentation/foo.html
# Documentation/gitignore.html
# file.o
# lib.a
# src/internal.o
[...]
$ cat .git/info/exclude
# ignore objects and archives, anywhere in the tree.
*.[oa]
$ cat Documentation/.gitignore
# ignore generated html files,
*.html
# except foo.html which is maintained by hand
!foo.html
$ git status
[...]
# Untracked files:
[...]
# Documentation/foo.html
[...]
Generally there are three different ways to ignore untracked files.
1) Ignore for all users of the repository:
Add a file named .gitignore to the root of your working copy.
Edit .gitignore to match your preferences for which files should/shouldn't be ignored.
git add .gitignore
and commit when you're done.
2) Ignore for only your copy of the repository:
Add/Edit the file $GIT_DIR/info/exclude in your working copy, with your preferred patterns.
Ex: My working copy is ~/src/project1 so I would edit ~/src/project1/.git/info/exclude
You're done!
3) Ignore in all situations, on your system:
Global ignore patterns for your system can go in a file named what ever you wish.
Mine personally is called ~/.gitglobalignore
I can then let git know of this file by editing my ~/.gitconfig file with the following line:
core.excludesfile = ~/.gitglobalignore
You're done!
I find the gitignore [2] man page to be the best resource for more information.
[1] http://www.kernel.org/pub/software/scm/git/docs/gitignore.htmlHow do you 'mark' 'tag' or 'release' a particular set of revisions for a particular set of files so you can always pull that one later?
Using the git tag
command.
To simply "tag" the current revision, you would just run..
git tag -a thetagname
git tag -a 0.1
git tag -a 2.6.1-rc1 -m 'Released on 01/02/03'
To list the current tags, simply run git tag
with no arguments, or -l
(lower case L):
$ git tag -a thetagname # and enter a message, or use -m 'My tag annotation'
$ git tag -l
thetagname
To delete a tag, you use the -d
flag:
$ git tag -d thetagname
Deleted tag 'thetagname'
$ git tag
[no output]
To tag a specific (previous) commit, you simply do..
git tag [tag name] [revision SHA1 hash]
For example:
git tag 1.1.1 81b15a68c6c3e71f72e766931df4e6499990385b
Note: by default, git creates a "lightweight" tag (basically a reference to a specific revision). The "right" way is to use the -a
flag. This will launch your editor asking for a tag message (identical to asking for a commit message, you can also use the -m
flag to supply the tag message on the command line). Using an annotated tag creates an object with its own ID, date, tagger (author), and optionally a GPG signature (using the -s
tag).
For further information on this, see
this post
[1]
git tag mytagwithmsg -a -m 'This is a tag, with message'
And to list the tags with annotations, use the -n1
flag to show 1 line of each tag message (-n245
to show the first 245 lines of each annotation, and so on):
$ git tag -l -n1
mytagwithmsg This is a tag, with message
For more information, see the git-tag(1) Manual Page [2]
[1] http://www.rockstarprogrammer.org/post/2008/oct/16/git-tag-does-wrong-thing-default/git tag tagname revision_SHA1
- dbr
Workflow example with GIT.
Git is extremely flexible and adapts good to any workflow, but not enforcing a particular workflow might have the negative effect of making it hard to understand what you can do with git beyond the linear "backup" workflow, and how useful branching can be for example.
This blog post [1] explains nicely a very simple but effective workflow that is really easy to setup using git.
quoting from the blog post: We consider origin/master to be the main branch where the source code of HEAD always reflects a production-ready state:
The workflow has become popular enough to have made a project that implements this workflow: git-flow [2]
Nice illustration of a simple workflow, where you make all your changes in develop, and only push to master when the code is in a production state:
Now let's say you want to work on a new feature, or on refactoring a module. You could create a new branch, what we could call a "feature" branch, something that will take some time and might break some code. Once your feature is "stable enough" and want to move it "closer" to production, you merge your feature branch into develop. When all the bugs are sorted out after the merge and your code passes all tests rock solid, you push your changes into master.
During all this process, you find a terrible security bug, that has to be fixed right away. You could have a branch called hotfixes, that make changes that are pushed quicker back into production than the normal "develop" branch.
Here you have an illustration of how this feature/hotfix/develop/production workflow might look like (well explained in the blog post, and I repeat, the blog post explains the whole process in a lot more detail and a lot better than I do.
[1] http://nvie.com/posts/a-successful-git-branching-model/Here's a copy of PJ Hyett's post, as it is not available anymore:
Git Isn't Hard
Nov 23, 2008
When we tell people why they should use Git over Subversion, the go-to line is, “Git does Subversion better than Subversion, but it does a lot more than that.”
The “lot more” is comprised of a bunch of stuff that makes Git really shine, but it can be pretty overwhelming for those coming from other SCM’s like Subversion.
That said, there’s nothing stopping you from using Git just like you use Subversion while you’re making the transition.
Assuming you’ve installed the necessary software and have a remote repository somewhere, this is how you would grab the code and push your changes back with Subversion:
$ svn checkout svn://foo.googlecode.com/svn/trunk foo
# make your changes
$ svn commit -m "my first commit"
And how would you do it in Git:
$ git clone git@github.com:pjhyett/foo.git
# make your changes
$ git commit -a -m "my first commit"
$ git push
One more command to make it happen in Git. That extra command has large implications, but for the purposes of this post, that’s all we’re talking about, one extra command.
See, it really isn’t that hard.
Update: I’d be remiss to not also mention that the equivalent of updating your local copy in Subversion compared to Git is
svn update
andgit pull
, respectively. Only one command in both cases.
Install msysgit [1]
There are several downloads:
This also installs a Cygwin bash shell, so you can use the git
in a nicer shell (than cmd.exe), and also includes git-gui (accessible via git gui
command, or the Start > All Programs > Git
menu)
Use the git-osx-installer [2], or you can also install from source
Install git
using your native package manager. For example, on Debian (or Ubuntu):
apt-get install git-core
Or on Mac OS X, via MacPorts [3]:
sudo port install git-core+bash_completion+doc
…or fink:
fink install git
…or Homebrew [4]:
brew install git
On Red Hat based distributions, such as Fedora:
yum install git
In Cygwin the Git package can be found under the "devel" section
In Mac OS X, if you have the Developer Tools installed, you can compile Git from source very easily. Download the latest version of Git as a .tar.bz
or .tar.gz
from http://git-scm.com/, and extract it (double click in Finder)
On Linux/BSD/etc. it should be much the same. For example, in Debian (and Ubuntu), you need to install the build-essential
package via apt
.
Then in a Terminal, cd
to where you extracted the files (Running cd ~/Downloads/git*/
should work), and then run..
./configure && make && sudo make install
This will install Git into the default place (/usr/local
- so git
will be in /usr/local/bin/git
)
It will prompt you to enter your password (for sudo
), this is so it can write to the /usr/local/
directory, which can only be accessed by the "root" user so sudo is required!
If you with to install it somewhere separate (so Git's files aren't mixed in with other tools), use --prefix
with the configure command:
./configure --prefix=/usr/local/gitpath
make
sudo make install
This will install the git
binary into /usr/local/bin/gitpath/bin/git
- so you don't have to type that every time you, you should add into your $PATH
by adding the following line into your ~/.profile
:
export PATH="${PATH}:/usr/local/bin/gitpath/bin/"
If you do not have sudo access, you can use --prefix=/Users/myusername/bin
and install into your home directory. Remember to add ~/bin/
to $PATH
The script x-git-update-to-latest-version [5] automates a lot of this:
[1] http://code.google.com/p/msysgit/This script updates my local clone of the git repo (localy at
~/work/track/git
), and then configures, installs (at/usr/local/git
-git describe
) and updates the/usr/local/git
symlink.This way, I can have
/usr/local/git/bin
in myPATH
and I'm always using the latest version.The latest version of this script also installs the man pages. You need to tweak your
MANPATH
to include the/usr/local/git/share/man
directory.
yum install git
. For the GUI run yum install git-gui
. - Cristian Ciupitu
sudo port install git-core+bash_completion+doc
- Singletoned
fink self-update
should help - dbr
Say you make a pull, merge it into your code, and decide you don't like it. Use git-log, or tig, and find the hash of wherever you want to go back to (probably your last commit before the pull/merge) copy the hash, and do:
# Revert to a previous commit by hash:
git-reset --hard <hash>
Instead of the hash, you can use HEAD^ as a shortcut for the previous commit.
# Revert to previous commit:
git-reset --hard HEAD^
git reset
should unstage accidental git add
- slf
How to set up a normal repository is described here [1] -- but how do you set up a team repository that everybody can pull and push from and to?
Assuming your team already has for instance a shared group membership that can be used.
mkdir /your/share/folder/project.git
cd /your/share/folder/project.git
newgrp yourteamgroup # if necessary
git init --bare --shared
To start using this repository the easiest thing to do is start from a local repository you already have been using:
cd your/local/workspace/project
git remote add origin /your/share/folder/project.git
git push origin master
Others can now clone this and start working:
cd your/local/workspace
git clone /your/share/folder/project.git
Set up a user account on the target server. Whether you use an account with no password, an account with a password, or use authorized_keys
really depend on your required level of security. Take a look at
Configuring Git over SSH
[2] for some more information.
If all developers use the same account for accessing this shared repository, you do not need to use the --shared
option as above.
After initing the repository in the same way as above, you do the initial push like this:
cd your/local/workspace/project
git remote add origin user@server:/path/to/project.git
git push origin master
See the similarity with the above? The only thing that might happen in addition is SSH asking for a password if the account has a password. If you get this prompt on an account without a password the SSH server probably has disabled PermitEmptyPasswords
.
Cloning now looks like this:
cd your/local/workspace
git clone user@server:/path/to/project.git
[1] https://stackoverflow.com/questions/315911/git-for-beginners-the-definitive-practical-guide/320140#320140git status
is your friend, use it often. Good for answering questions like:
Unlike, say svn status
, git status
runs nigh-instantly even on large projects. I often found it reassuring while learning git to use it frequently, to make sure my mental model of what was going on was accurate. Now I mostly just use it to remind myself what I've changed since my last commit.
Obviously, it's much more useful if your .gitignore is sanely configured.
Once you've edited a file, you need to commit your changes to git. When you execute this command it will ask for a commit message - which is just a simple bit of text that tells everyone what you've changed.
$ git commit source/main.c
Will commit the file main.c in the directory ./source/
$ git commit -a # the -a flag pulls in all modified files
will commit all changed files (but not new files, those need to be added to the index with git-add). If you want to commit only certain files then you will need to stage them first with git-add and then commit without the -a flag.
Commiting only changes your local repository though not the remote repositories. If you want to send the commits to the remote repository then you will need to do a push.
$ git push <remote> <branch> # push new commits to the <branch> on the <remote> repository
For someone coming from CVS or SVN this is a change since the commit to the central repository now requires two steps.
The default branch in a git repository is called master
.
To create a new branch use
git branch <branch-name>
To see a list of all branches in the current repository type
git branch
If you want to switch to another branch you can use
git checkout <branch-name>
To create a new branch and switch to it in one step
git checkout -b <branch-name>
To delete a branch, use
git branch -d <branch-name>
To create a branch with the changes from the current branch, do
git stash
git stash branch <branch-name>
$ git pull <remote> <branch> # fetches the code and merges it into
# your working directory
$ git fetch <remote> <branch> # fetches the code but does not merge
# it into your working directory
$ git pull --tag <remote> <branch> # same as above but fetch tags as well
$ git fetch --tag <remote> <branch> # you get the idea
That pretty much covers every case for getting the latest copy of the code from the remote repository.
The Pro Git [1] free book is definitely my favorite, especially for beginners.
[1] http://progit.orgGit Magic [1] is all you'll ever need. Guaranteed or your money back!
[1] http://www-cs-students.stanford.edu/~blynn/gitmagic/If you want to merge a branch (e.g. master
to release
), make sure your current branch is the target branch you'd like to merge into (use git branch
or git status
to see your current branch).
Then use
git merge master
(where master
is the name of the branch you want to merge with the current branch).
If there are any conflicts, you can use
git diff
to see pending conflicts you have to resolve.
I've also found Git Internals [1] to be very useful. It is written by Scott Chacon (author of Pro Git, and maintainer of the Git Community Book). What I like about Git Internals is it focuses on the concepts first and then the commands [2], and being that it is ~100 small pages it is quickly digestible.
[1] http://peepcode.com/products/git-internals-pdfgit log -- filename
Assuming there is a remote repository that you cloned your local repository from and also assuming that there is a branch named 'some_branch' on that remote repository, here is how to track it locally:
# list remote branches
git branch -r
# start tracking one remote branch
git branch --track some_branch origin/some_branch
# change to the branch locally
git checkout some_branch
# make changes and commit them locally
....
# push your changes to the remote repository:
git push
git remote show REMOTENAME
- Felipe Sabino
A real good paper for understanding how Git works is The Git Parable [1]. Very recommended!
[1] http://tom.preston-werner.com/2009/05/19/the-git-parable.htmlCompare command is git diff
.
To compare 2 revisions of a file:
$ git diff <commit1> <commit2> <file_name>
That diffs commit1 against commit2; if you change order then files are diffed the other way round, which may not be what you expect...
To compare current staged file against the repository:
$ git diff --staged <file_name>
To compare current unstaged file against the repository:
$ git diff <file_name>
Why yet another howto? There are really good ones on the net, like the git guide [1] which is perfect to begin. It has good links including the git book [2] to which one can contribute (hosted on git hub) and which is perfect for this collective task.
On stackoverflow, I would really prefer to see your favorite tricks !
Mine, which I discovered only lately, is git stash
, explained
here
[3], which enables you to save your current job and go to another branch
EDIT: as the previous post, if you really prefer stackoverlow format with posts as a wiki I will delete this answer
[1] http://www.sourcemage.org/Git_Guideapt-get install tig
While inside a git repo, type 'tig', to view an interactive log, hit 'enter' on any log to see more information about it. h for help, which lists the basic functionality.
"Tig" is "Git" backwards.
I got started with the official Git tutorial [1]. I think it's practical enough for beginners (I was, and still am, a beginner, by your definition! I barely grasp makefiles, I've only played a bit with Apache Subversion, etc.).
[1] http://www.kernel.org/pub/software/scm/git/docs/gittutorial.htmlAssuming that you have cloned your remote repository from some single remote repository.
# create a new branch locally
git branch name_of_branch
git checkout name_of_branch
# edit/add/remove files
# ...
# Commit your changes locally
git add fileName
git commit -m Message
# push changes and new branch to remote repository:
git push origin name_of_branch:name_of_branch
git push origin name_of_branch
and the branch will already be created in your remote - Felipe Sabino
name_of_branch
is the local name, the second is the (desired) remote branch name, so it could be local_name_of_branch:remote_name_of_branch
if you want the names to differ. If you want them the same, you still have to specify it like this b/c git doesn't make the assumption that you want the name to be the same unless you tell it so (there are other methods to do so as well, however) - johnny
Perform a push in your remote using :
before the name of the branch
git push origin :mybranchname
being origin
the name of your remote and mybranchname
the name of the branch about to be deleted
http://help.github.com/remotes/
Push and pull changes
In an simplified way, just do git push
and git pull
. Changes are merged and if there's a conflict git will let you know and you can resolve it manually.
When you first push to a remote repository you need to do a git push origin master
(master being the master branch). From then on you just do the git push
.
Push tags with git push --tags
.
First go to an empty dir, use "git init" to make it a repository, then clone the remote repo into your own.
git clone user@host.com:/dir/to/repo
Wherever you initially clone from is where "git pull" will pull from by default.
Resource: Definitely check out Scott Chacon's Gitcasts [1], especially the Railsconf talk [2].
Github [3] is awesome and also has some helpful guides [4].
[1] http://www.gitcasts.com/WRT good GUIs/frontends, you may also want to check out qgit [1] which is a cross-platform (Linux/Win32) repository viewer for Git and can be also used as high level frontend for the most common Git operations, in fact it can be easily enhanced by so called "custom actions" so that users can provide customized actions.
[1] http://sourceforge.net/projects/qgitSeriously add the link featured in Tim's answer [1] in the Stack Overflow post Setup Git Server with Msysgit on Windows [2].
It flawlessly told me how to setup Git on Windows with msysgit [3], and is an incredibly detailed article.
[1] https://stackoverflow.com/questions/1482824/setup-git-server-with-msysgit-on-windows/1503986#1503986Rebase explanation taken from the book Pragmatic Guide to Git - Travis Swicegood [1]
[1] https://rads.stackoverflow.com/amzn/click/com/1934356727Chapter III
16 . Rewriting History by Rebasing
Rebasing commits is the one concept in Git that has no counterpart inside the traditional version control world. Using git rebase, you can rewrite the history of a repository in a variety of ways. It is one of the most powerful commands in Git, which makes it one of the most dangerous.
rebase
takes a series of commits (normally a branch) and replays them on top of another commit (normally the last commit in another branch). The parent commit changes so all the commit IDs are recalculated. This can cause problems for other developers who have your code because the IDs don’t match up.There’s a simple rule of thumb with git rebase: use it as much as you want on local commits. Once you’ve shared changes with another developer, the headache is generally not worth the trouble.
git rebase --abort
. Everything will be back to the way it was before you started. And: Don't know if you want to mention this here (it seems less beginner-y), but you can also rebase --onto
. - torek
I found this post [1] to be very useful to get me started. I still need to read the book and other resources but the post was helpful in, as the title says, "understanding git conceptually". I also recommend taking the Git & GitHub course offered at RubyLearning [2].
[1] http://www.eecs.harvard.edu/~cduan/technical/git/One more item I really think should be in this list, probably very useful for beginners:
What if I've done some commits and then I did something scary, like maybe a rebase, and now something—or even everything—seems to be lost? (Rebase seems to be the one that gets most people the first time, so I'm concentrating on it. While git rebase --abort
helps a lot, sometimes you'll find that you botched an edit during an interactive rebase, for instance, and let the rebase finish and now you want to get your old stuff back. And then there are things like filter-branch
...)
One key git principle is that it never actually deletes anything you've committed. ("What, never?" "No, never!" "What, never?" "Well, hardly ever!") If you have not run git gc
, it's still in there. It may take some digging around to find your previous work, but if you did some successful git commit
s earlier, then, for instance, even your apparently-wrecked series of commits from a tragic rebase error are still in there, normally for at least a month (technically, until the "reflogs" expire).
It's important to keep in mind that each branch name labels—or points to—a "commit-ID". These are the the funny numbers like 7cc5272
. Many of the things you do, like adding a new commit to a branch, make the branch name point to a new, different commit-ID. Each commit-ID has a link pointing back to some previous commit-ID(s), and this is what actually makes up a "branch" full of commits.
The rebase entry talks about "rewriting history," and commands like git filter-branch
also "rewrite history," but they do it not by destroying the previous history, but rather by adding new history. Once the new history is in place, git will "move the labels around" so that it looks like history has changed. If you are on your fix-nasty-bug
branch and do a git rebase
and manage to wreck things, the label fix-nasty-bug
now refers to the wreckage, but the original versions are still there. Rebase in particular makes a temporary (non-moving, not-a-branch) label spelled ORIG_HEAD
that lets you find them. The filter-branch
command saves all the original names as well. In some cases, there may be no obvious name, but the commits can always be found. If necessary, find yourself a "git guru" and explain what you did that led to the wreckage.
(The command git reflog show
can also help with finding commit-IDs.)
If you have found what you think is some or all of your previous work, try:
git log <commit-ID> # ORIG_HEAD after a bad rebase, for instance
git show <commit-ID> # or some long SHA1 value you can still see in a window
If it looks right or useful, put a name to it:
git branch recover-my-stuff ORIG_HEAD
and it's all back again! In fact, now both your bad rebase and your original work are in your git repo "forever" (or at least, until you delete the branch names and let a few months go by, and then they get garbage-collected). You can put as many names to as many recovered commits as you like. (Branch names are virtually free, except for cluttering up your git branch
output, and of course they also keep commits from being garbage-collected. You can also, or instead, put tags on specific commit-IDs, if you prefer those.)
Very good post on merging with conflicts - GitGuys: Merging With a Conflict - Conflicts And Resolutions [1]
The blog is really great - illustrative, clean examples and understandable. Definitely worth checking out.
[1] http://www.gitguys.com/topics/merging-with-a-conflict-conflicts-and-resolutions/
rm -rf .git
to obliterate git's database. - Josh Leegit status
to ensure you're within a repository: this has one gotcha: if you have the environment variable $GIT_DIR set in your current shell, git will ignore your current location and use the repository at $GIT_DIR. I should know, I lost an hour to that yesterday. - sanmiguel